删除重复记录,将TABLE_NAME中的不重复记录保存到#TABLE_NAME中
select distinct * into #table_name from table_name
delete from table_name
select * into table_name from #table_name
drop table #table_name
与此相关的是“select into”选项,可以在数据库属性
对话框中,勾起来此项,或者在Query Analyzer中执行
execute sp_dboption 'db_name','select into','true'
开启。默认值是关闭的。
*******************************************************
收缩事务日志(多次执行)
backup log register with NO_LOG
backup log register with TRUNCATE_ONLY
DBCC SHRINKDATABASE(register)
更多有用的sql语句
/*sql 语法学习*/
/*函数的学习---------------------------------------*/
获取当前时间(时/分/秒):select convert(varchar(10),getdate(),8)
获取当前年月日:select convert(varchar(10),getdate(),120)
获取当前年月:select convert(varchar(7),getdate(),120)
获取当前年月:select convert(varchar(10),year(getdate())) + '-' + convert(varchar(10),month(getDate()))
select cast(b as integer) as bb from table1 where b = '11'
select a,case b when '11' then '细细' when '22' then '呵呵' else '哈哈' end as 转换,c from table1
select a,b,case when c = '111' then '细细' when c = '222' then '呵呵' else '哈哈' end as 转换1 from table1
获取当前时间:print current_timestamp
/*---------------------------------------------*/
-----------------将sql查询输出到txt文本文件中-------------------------------------------
EXEC master..xp_cmdshell 'bcp 数据库名.dbo.表名 out d:\1.txt -c -q -U"sa" -P"password"'
---------------------------------------------------------------------------------------
---------------------------round的用法beigin------------------------------
declare @s float
set @s = 0.1566134
print round(@s,3)
---------------------------round的用法end---------------------------------
--------------------------------自动收缩数据库begin-----------------------------
EXEC [master]..sp_dboption [Database Name], 'autoshrink', 'TRUE'
--------------------------------自动收缩数据库end-----------------------------
-------------------------------去除首尾无效的字符begin--------------------------
declare @s varchar(20)
set @s=',,,1->1,'
while(left(@s,1)=',')
set @s=stuff(@s,1,1,'')
while(right(@s,1)=',')
set @s=stuff(reverse(@s),1,1,'')
select @s
-------------------------------去除首尾无效的字符end--------------------------
------------删除数据库中的重复记录(且仅保留一条有效记录)示例-----------------
create table A
(
userID int identity(1,1),
userName varchar(20),
userPwd varchar(20),
userEmail varchar(50)
)
insert into A(userName,userpwd) select 'qin','qin' union all select 'qin','qin1' union all select 'qin','qin1'
select * from A
--method one
delete from A where userid not in(select min(userid) as userid from A group by username ,userpwd)
--method two
delete from A where exists (select * from A b where a.username = b.username and a.userpwd = b.userpwd and a.userid < b.userid)
--method three
delete from a where userid not in(select min(userid) from A b where a.username = b.username and a.userpwd = b.userpwd and a.userid > b.userID)
select * from A
drop table A
------------删除数据库中的重复记录(且仅保留一条有效记录)示例-----------------
-------------------------------迭归的应用(找起点和终点之间的路径-----------------------------
create table t
(st varchar(20),ed varchar(20),km int)
go
insert t values ('A','B',1000)
insert t values ('A','C',1100)
insert t values ('A','D',900)
insert t values ('A','E',400)
insert t values ('B','D',300)
insert t values ('D','F',600)
insert t values ('E','A',400)
insert t values ('F','G',1000)
insert t values ('C','B',600)
go
--显示插入值
select * from t
go
--创建函数
--函数返回一个表,根据实际情况的不同一层一层的插入,可以充分利用生成的表
create function f_go(@col varchar(10))
returns @t table(col varchar(30),st varchar(20),ed varchar(20),km int,level int)
as
begin
declare @i int
set @i=1
insert @t select st+'-'+ed,*,@i from t where st=@col
while exists (select * from t a,@t b where
b.ed=a.st and b.level=@i and b.ed<>@col )
begin
set @i=@i+1
insert @t
select b.col+'-'+a.ed,a.st,a.ed,b.km+a.km,@i from t a,@t b
where b.level=@i-1 and b.ed=a.st and b.ed<>@col
end
return
end
go
--调用
--select * from dbo.f_go('A')
select col,km from dbo.f_go('a')
--删除环境
drop function f_go
drop table t
-------------------------------迭归的应用(找起点和终点之间的路径-----------------------------
--------按类别去最新的前N条记录,把同一类的放在一起,统计同一类的项的个数等-------------
create table t
(
ClassName varchar(50),
ClassCode varchar(10),
ClassID int identity(1,1)
)
insert into t
select 'cccc1','002' union all
select 'aaaa','001' union all
select 'bbbb','001' union all
select 'aaaa1','002' union all
select 'cccc','001' union all
select 'dddd','001' union all
select 'bbbb1','002' union all
select 'dddd1','002'
select * from t
select ClassCode = (case when exists(select 1 from t t1 where classCode = t1.ClassCode
and ClassID < t1.ClassID)
then '' else ClassCode end),ClassName from t order by ClassCode,ClassID desc
select count(*),classCode from (select top 100 percent ClassCode = (case when exists(select 1 from t t1 where classCode = t1.ClassCode
and ClassID < t1.ClassID)
then '' else ClassCode end),ClassName from t order by ClassCode,ClassID desc)a group by classcode
select classCode,className from t order by classCode,classID desc
drop table t
--------按类别去最新的前N条记录,把同一类的放在一起,统计同一类的项的个数等-------------
-------------同上,按类别进行统计,把同一类的项的其他内容进行相加并发在一个字段中------------------
create table tb(ProductID varchar(10),PositionID varchar(10))
insert into tb
select '10001','A1'
union all select '10001','B2'
union all select '10002','C3'
union all select '10002','D4'
union all select '10002','E5'
go
create function dbo.fc_str(@ProductID varchar(10))
returns varchar(100)
as
begin
declare @sql varchar(1000)
set @sql=''
select @sql=@sql+','+cast(PositionID as varchar(20)) from tb where ProductID=@ProductID
return stuff(@sql,1,1,'')
end
go
select ProductID,dbo.fc_str(ProductID) as PositionID from tb group by ProductID
drop table tb
drop function dbo.fc_str
-------------按类别进行统计,把同一类的项的其他内容进行相加并发在一个字段中------------------
--取各个类的前n条记录(每个类都取top n条)
--如果有数据库中有多个类,现在要取每个类的前n条记录,可用以下语句
Create Table TEST
(ID Int Identity(1,1),
h_id Int)
Insert TEST Select 100
Union All Select 100
Union All Select 100
Union All Select 101
Union All Select 101
Union All Select 101
Union All Select 100
GO
--方法一:
Select * From TEST A Where Id In(Select TOP 3 ID From TEST Where h_id=A.h_id)
--方法二:
Select * From TEST A Where Not Exists (Select 1 From TEST Where h_id=A.h_id And ID<A.ID Having Count(*)>2)
--方法三:
Select * From TEST A Where (Select Count(*) From TEST Where h_id=A.h_id And ID<A.ID)<3
GO
Drop Table TEST
GO
--分组统计,统计每个段中数据的个数
--一般成绩统计可以用到这个
declare @t table(id int,weight int)
insert into @t select 1, 20
insert into @t select 2, 15
insert into @t select 3, 5
insert into @t select 4, 60
insert into @t select 5, 12
insert into @t select 6, 33
insert into @t select 7, 45
insert into @t select 8, 59
insert into @t select 9, 89
insert into @t select 10,110
declare @p int
set @p=10
select
rtrim(p*@p)+'-'+rtrim((p+1)*@p">p*@p)+'-'+rtrim((p+1)*@p) as p,
num
from
(select (weight/@p">weight/@p) as p,count(*) as num from @t where weight between 10 and 100 group by (weight/@p">weight/@p)) a
----------------------------在in语句中只用自定义排序begin--------------------------------
declare @t table(id int,weight int)
insert into @t select 1, 20
insert into @t select 2, 15
insert into @t select 3, 5
insert into @t select 4, 60
insert into @t select 5, 12
insert into @t select 6, 33
insert into @t select 7, 45
insert into @t select 8, 59
insert into @t select 9, 89
insert into @t select 10,110
--默认in语句中sql会按照id进行排序
select * from @t where id in(2,4,3)
--用此方法可以按照我们传入的id顺序进行显示数据
select * from @t where id in(2,4,3) order by charindex(rtrim(id),',2,4,3,')
----------------------------在in语句中只用自定义排序end--------------------------------